Summary of Key Points
This news report discusses the recent developments in the global AI industry: the global AI giant OpenAI has suddenly made its more powerful GPT-5.6 Luna model available to free users, offering unlimited chat capabilities, while the domestic AI company DeepSeek has announced significant price increases. Behind these moves lies a competitive landscape where domestic open-source AI models have attracted OpenAI's developer users with their lower prices, prompting OpenAI to adopt a free strategy in response. However, domestic models are also facing financial losses due to their low pricing and have had to shift from a loss-making model to one that charges based on the value provided to users. These changes are reshaping the global AI pricing landscape and sparking debates about whether open-source or closed-source approaches are more beneficial for AI development.
OpenAI's Free Strategy: A Defensive Countermeasure
OpenAI's decision to make GPT-5.6 Luna free is not a gesture of generosity but a response to the competition from domestic models:
- Initial Price Cuts Were Ineffective: A week ago, OpenAI reduced the price of GPT-5.6 Luna by 80% (from $6 per million tokens to $1.2), but developers still switched to other models. Third-party data shows that the usage of DeepSeek V4 Flash was three times that of Luna.
- Free as a Last Resort: OpenAI has now made Luna available for free users, adding a “Think” button that allows them to access advanced reasoning functions. Paid users can use the more sophisticated Sol model, which unifies both quick responses and in-depth thinking capabilities.
- Clear Purpose: The goal is to attract a large number of users and collect data using the free model while generating revenue from the paid model. This indicates that domestic models have threatened OpenAI's pricing power; OpenAI must now rely on free services to retain its user base.
The Counterattack of Domestic Open-Source Models: Winning the Market with Low Prices, but at a Cost
Domestic open-source models like DeepSeek were able to gain momentum due to their affordable prices and performance. However, they are now facing profitability challenges:
- Low Prices Attract Users: DeepSeek V4 Flash costs only 35% of Luna’s price ($0.06 per million tokens) while offering similar performance, attracting many developers. The OpenCode platform processes 8 trillion tokens daily.
- Financial Losses: Running AI models requires substantial computing power, and the energy consumption for intelligent applications (such as task automation) is 100 times that of regular conversations. The more models are used, the greater the losses, leading to issues like server capacity shortages during peak times.
- Rising Prices Necessary: DeepSeek’s price increase reflects a broader industry trend. Similar increases have been seen with platforms like Zhipu API (83% rise in prices) and cloud services from Alibaba Cloud and Tencent Cloud. This shows that users are willing to pay for stable and reliable services, prompting model manufacturers to shift from low-price strategies to value-based pricing.
The Change in AI Pricing Logic: From Low Prices to Value-Based Pricing
Previously, AI model manufacturers focused on attracting users with low prices. However, they have realized that losing money while offering free services is not sustainable, so they are moving towards value-based pricing:
- What is Value-Based Pricing?: It means charging based on the value that the model provides, rather than just the amount of text processed. For example, despite the price increase, users continue to use Zhipu API because it helps companies save costs.
- Reasons for the Change: The cost of computing power has risen significantly. Training and running large models requires many GPUs, which are expensive in terms of electricity and server costs. Maintaining low prices would be unsustainable for manufacturers.
- Impact on Users: AI services will likely become less affordable, but users will receive more stable and intelligent solutions. For instance, DeepSeek’s price increase should result in improved service stability.
The Debate over Open Source and Closed Source: Which Path Leads to Better AI Development
This incident highlights the differing perspectives within the AI industry:
- Open Source Advocates: Leaders like Hugging Face’s CEO argue that open-source technologies could have led to greater progress if not for companies hiding them for profit. Turing Award winner LeCun also supports open source, stating that AI cannot thrive in a closed environment (he left Meta because of its increasingly closed approach).
- Closed Source Supporters: OpenAI initially used open source but later switched to a closed model for profitability. They argue that closed models protect technical secrets and generate more revenue but may hinder industry progress.
- Advantages of Domestic Models: Many domestic models are open-source, allowing developers to modify and improve them, which has contributed to their rapid development and market success.
The Global AI Landscape: Changing Rules with the Rise of Domestic Models
The rise of domestic open-source models is changing the global AI landscape:
- No Longer Monopolized by Silicon Valley: OpenAI no longer has complete control over pricing. It must now consider the competition from domestic models, adjusting its strategies accordingly.
- Competition Shifts to Ecosystems: Success will depend on building more open, affordable, and stable ecosystems. Domestic models, with their open-source approach and low prices, have already established a foothold in the developer community.
- Future Trends: The AI industry will become more competitive, but users will benefit from more affordable and advanced services. Manufacturers will need to find a balance between profitability and development.
In summary, OpenAI’s free strategy and DeepSeek’s price increase are not isolated events; they mark a transition from the wild growth of the AI industry to a more rational approach. The rise of domestic models is shifting the balance of power in the industry, with users ultimately benefiting from more affordable and efficient AI services.